A Monocular Camera Internal Parameters Calibration Height Invariance Method For Smart Agriculture

Technology is the fundamental driving force behind building a strong agricultural nation.The concept of smart agriculture has been vigorously promoted in recent years and has become one of the main directions for future agricultural development.Agricultural robots are a core part of realising smart agriculture,and the distance information in the scene is an important basis for intelligent decision making by agricultural robots.The monocular vision-based ranging method can simultaneously achieve target detection,identification and ranging localization with low equipment cost and model complexity,so it has high research application value.Most of the current ranging methods based on geometric ranging models do not take into account the influence of camera nonlinear imaging on the model,and it is necessary to know the height of the measured object or the distance between the feature points in the process of ranging,which cannot meet the high accuracy of ranging any target in both static and dynamic scenes,so their application value is limited.To meet the high accuracy of the ranging model in static and dynamic scenes,this thesis proposes a monocular visual ranging model that can be applied under dynamic changes in camera height to meet the real-time accurate ranging of livestock and crop targets with contact points with the ground in agricultural operation scenarios.The specific work of this thesis is as follows:1.This thesis introduces a monocular visual ranging model based on the principle of keyhole imaging.The focal length invariance of the ranging model under the condition of camera height change is proved by the model derivation and hypothesis testing,and verification experiments are conducted from two perspectives: a set of focal length regression vectors are jointly calibrated with pixel points captured at multiple camera heights and a set of focal length regression vectors are used to achieve multiple shot height ranging.The experimental results demonstrate that the minimum ranging accuracy of 95.8% and the average ranging accuracy of 96.5% are achieved by using only one set of focal regression vectors to complete multiple height ranging.The results show that the camera only needs one calibration to achieve full height applicability.The complexity of focal length calibration is greatly reduced,allowing the model to be used for ranging in dynamic scenes.2.It is demonstrated that the focal length regression vector obtained from the calibration on the right side of the image plane y-axis can be used to calculate the focal length of the imaging point on the left side of the y-axis,resulting in a maximum loss of0.8% in range accuracy.3.The radial distortion is combined with the ranging model under height variation,and the radial distortion is used to correct the measurement error caused by the camera height variation.In this thesis,the segmentation curve function is determined by the fitting method,and the measured points that need radial distortion reduction are identified.The experimental results show that the maximum ranging accuracy of a single point can be increased by 1.93% and the minimum ranging accuracy can reach 96.46% by restoring the radial distortion of the physical coordinates of the measured points on the outside of the segmentation curve.The experimental verification is also carried out after the camera Angle and shooting height are changed.The results show that the maximum ranging accuracy of a single point increases by 2.25% and the minimum ranging accuracy reaches97.91% after the camera attitude changes.Several sets of experiments have verified the effectiveness of the correction method proposed in this thesis and its migration in the case of camera pose change.